Slide 15 of 21
Notes:
Pacific Ocean, June 12, 1996 -- Marines from the 11th MEU board a CH-46 Sea Knight Helicopter aboard USS Essex for a helo raid on Marine Corps Base Hawaii. (Official U.S. Navy Photo by JO2 Dave Kistler)
Very expensive task maintaining the transmissions of Marine Corps helicopters. It takes a long time and a lot of effort to dismantle the transmission, inspect it and reassemble it. On the other hand, a failure to identify faults in the system is very likely to be fatal. Current solution is scheduled maintenance.
Marines were very interested in developing a system that could detect problems with the transmission before they became dangerous. The task also poses problems to the collection of data. Ideal data are collected from flying with a faulty transmission, but that is both expensive and dangerous. Standard neural networks need many examples of faulty transmissions.
The Gluck model is ideally suited to detection of faults, because it learns the pattern of normal operation from operating transmissions and signals when the transmission operates in a nonpredicted (novel) manner. During testing the model on 9 examples of presumably operational helicopters, the network signalled that one was operating contrary to prediction. This copter was removed from service and was indeed found to be faulty.
